generative-adversarial-networks

experimenting with GANs on vast.ai

notebook contains code for training a GAN on aligned celeba images

architecture includes:

hinge loss
spectral norm on all layers
dense self attention
exponential moving average
two timescale update rule